To me, these examples vividly illustrate the limitations of working with observational data, and are an extension of the standard "correlation vs causation" pro

To me, these examples vividly illustrate the limitations of working with observational data, and are an extension of the standard "correlation vs causation" problem. I think the pertinent lesson here is that these are still very real problems in the world of "big data": just because you're map-reducing, and throwing around petabytes of data, what you can learn is still limited by the regime under which the data were collected.